The Next Phase of SAN Deployment: Storage Switch Technology
The architecture of open enterprise storage is undergoing a fundamental change. Today's model of distributed servers, each with their own private connection to storage, is rapidly transitioning to a network of storage resources shared by many heterogeneous servers. This new storage architecture has assumed its own identity as a storage area network (SAN). Fibre Channel, an open industry standard, has emerged as the technology foundation for the SAN because of its suitability for storage networking. Desirable Fibre Channel attributes include high bandwidth, long-distance connections, scalability, and broad industry support from major enterprise server, storage, and networking providers.
Modeling client/server LANs, the SAN environment interconnects servers and storage using a network of high-speed connections (see Figure 5)6. SAN eliminates the performance bottlenecks, distance constraints, and scalability limitations imposed by traditional SCSI-based architectures, enabling a more distributed storage network for enterprise environments.
Figure 5 SAN-LAN environment.
What's Driving SAN Evolution?
Initial deployments of SAN have focused on displacing SCSI interconnects mainly in single-server or two-node cluster environments where extended distances and higher storage scalability is required. Storage hubs, based on Fibre Channel technology, lower total cost of ownership by consolidating connections and enabling centralized management of the expanded storage resources. According to market research firm International Data Corporation, the total cost of managing storage can be reduced by 60% and IT administrators can effectively manage 970% more storage capacity under a centralized SAN environment, as compared to the traditional distributed server and storage model. In early deployments of SAN, storage hubs provide the simplest and most cost-effective solution (see Figure 6)6.
Figure 6 The storage hub replaces private SCSI interconnects.
Based on the promise of significantly lower total cost of ownership when multiple servers share a centralized pool of storage, the industry is now preparing to broaden the application and scope of the SAN to include more complex implementations, necessitating the addition of switching. As mainstream operating systems and storage devices support multiserver applications, these servers will require concurrent access to the shared storage resources. The switch provides this concurrency as well as necessary services such as zoning, the ability to segment the network and bind each server to specific disk or tape storage resources (see Figure 7)6. However, although the desire to introduce switching to SANs is quite clear, it is not without its challenges.
Figure 7 SAN switching benefit.
While all of the servers, hubs, and storage subsystems are based on the widely supported Fibre Channel protocol known as Fibre ChannelArbitrated Loop (FC-AL), the early switch products are based on a newer protocol called Fabric and are positioned more for SAN backbone applications and less for storage consolidation applications. Although they do provide the required concurrency and switch services, these Fabric switches have proven to be extremely problematic and costly to implement.
FC-BB defines backbone switches (BBWs) that connect SAN islands across WANs.
Much of this difficulty appears to be directly related to the complexity of the Fabric protocol itself. Attempts to achieve multivendor interoperability with backbone switches have proven to be a daunting challenge leading some to dub the Fabric protocol as a "nonstandard standard" (see Figure 8)6.
Figure 8 The problem with Fabric.
In addition to their complexity, the cost of backbone switches is a barrier to their adoption. Backbone switches are inherently expensive, typically four to eight times the cost per port of a storage hub. However, for many users and integrators, the transition and support costs are perhaps even more painful than the purchase price. Because the Fabric protocol does not directly connect to existing storage devices and host adapters, implementers must re-engineer the network with specialized software, firmware, and hardware on all of the host and storage connections. Extensive multivendor interoperability testing must also be performed to verify the stability and reliability of the Fabric-based network. These cost and complexity barriers have severely limited the addition of switching within the SAN. And until recently, SAN implementers had no alternative to the backbone switch.
The Storage Switch: A Simple, Cost-Effective Approach to SAN Switching
Fortunately, a new switch class was recently introduced to the SAN: the storage switch (or loop switches). The storage switch delivers the same concurrency and zoning services offered by the backbone switch, while eliminating the cost and complexity of a mixed-protocol environment (see Figure 9)6. How is this possible? Based on an innovative implementation of the existing industry standard FC-AL protocol, the storage switch maintains the same technology and characteristics of the existing installed base of storage hubs, storage subsystems, and servers. This approach challenges the common misconception that the FC-AL protocol is limited for use only in a shared-bandwidth loop topology or that the Fabric protocol is required for a switched topology. In truth, the FC-AL standard defines a simple and cost-effective communication protocol that is well suited for both loop and switch topologies. Embracing the FC-AL protocol, a storage switch eases the addition of switching and provides a flexible building block for continued SAN evolution and rapid deployment.
Figure 9 A storage switch provides the best of both worlds.
Applying a Storage Switch in the SAN: One-Tier SAN
For small networks with multiple servers, a storage switch can immediately boost storage performance by allowing concurrent traffic between each server and the shared storage resources (see Figure 10)6. Because the storage switch is based on the same protocol as the storage hub and other connected devices, the hub can literally be removed and replaced with a storage switch without any other changes to the network. This unique attribute of the storage switch drastically reduces the engineering effort and completely preserves the installed base of connected end-node devices. As the network grows, it's easy to add another storage switch or storage hub to facilitate more servers or expanded storage capacity.
Figure 10 A storage switch easily drops in place of a hub to improve performance.
In SAN environments where higher storage scalability is a priority, a combination of hubs and switches can be utilized to construct a two-tiered SAN (see Figure 11)6. In this environment, the storage hub facilitates the broad, concentrated connectivity for storage, while a storage switch takes the place of the backbone switch to provide the needed concurrency and zoning. This network design can support scaling storage capacity to hundreds or even thousands of terabytes using existing loop protocol standards and technology without ever introducing the Fabric protocol and deferring the need for a costly backbone switch.
Figure 11 A storage switch directly connects to existing storage hubs in a two-tier SAN.
In extremely large topologies where scalability to thousands or tens of thousands of terabytes is required, the backbone switch can be added at the top tier to provide the needed connectivity to thousands of host and storage devices. However, this introduces network complexities that must first be addressed. Because most backbone switches today support only 8 to 16 ports, building a network of this size would require a complex interswitch linking scheme and would likely introduce serious performance bottlenecks moving data from one end of the network to the other. These problems will certainly be solved in future phases of SAN deployment, and the storage switch will complement the existing storage hub and future backbone switch in a three-tier network hierarchy of SAN infrastructure components.
By enabling more graduated steps in multitier architecture, a storage switch flattens out the cost of ownership curve as node count increases by providing building blocks for optimal scalability. With a storage switch, scaling of storage to thousands of terabytes can be accomplished without ever introducing mixed protocols.
Although backbone switch vendors are adopting technology to allow direct connection to both FC-AL and Fabric-aware storage devices, it adds even more cost to the implementation (see Figure 12)6. In shedding the complexity of Fabric, the storage switch becomes a more cost-effective solution. When you consider the additional cost of transitioning to Fabric-aware host adapters and the cost of integration and support, the total cost of acquisition, integration, and support for the a storage switch drops even further.
Figure 12 Cost and scalability comparison.
Industry Validation Of The Storage Switch
Major system, storage, and networking leaders such as 3Com, Clariion, EMC, Hewlett-Packard, Hitachi Data Systems, Intel, and Seagate have endorsed the storage switch technology. These companies view the storage switch as a positive step for the industry that will accelerate the migration to switching, and they are anticipating availability of products beginning in early 2002. Simple implementation, investment protection, and the ability to apply previous knowledge and practices will accelerate the adoption of SAN switching technology. Fibre Channel Fabric switch benefits have been limited by the complexity, cost, and interoperability issues.
One thing is clear: The technology is based on existing industry standards, so SAN switch providers are free to implement a storage switch product today as an attractive alternative to the backbone switch.
Next, let's explore how deploying an enterprise storage area network (SAN) can help make the corporate information utility a reality.